Fragments: AI's Impact on Development, Governance, and Society

In the rapidly evolving landscape of software development, artificial intelligence is reshaping how we write, maintain, and think about code. This collection of observations explores the multifaceted impact of AI on programming practices, organizational policies, and societal attitudes.

Two Veterans on AI and Programming

At the GOTO Conference in Copenhagen in 2025, Martin Fowler and Kent Beck engaged in what Fowler describes as "two old geezers on a park bench" - a candid discussion about their experiences with LLM-augmented programming as of October 2025. Their conversation reveals a mix of frustration and optimism, noting that fundamental principles they've advocated for decades still need emphasis. Both agree that any manifesto-like movements need to be led by younger generations, and they offer perspectives on what junior developers should focus on in their careers.

Fowler reflects on the enduring nature of software development wisdom: "We show our frustration that things we've been saying for thirty years still need to be said." This highlights the challenge of establishing lasting practices in an industry constantly distracted by new technologies.

Restructuring Legacy Code with AI Assistance

Ian Johnson has documented an impressive journey of restructuring a gnarly Laravel + React codebase over approximately three months and 258 commits. The transformation took the application from a legacy monolith with no tests to a well-structured system with automated quality gates, a React SPA migration in progress, and an AI agent capable of reliably shipping production code with minimal supervision.

Johnson's approach follows established software engineering principles:

1. First, getting everything under the control of characterization tests
2. Adding static analysis
3. Introducing appropriate architectural patterns

His experience with AI evolved significantly during the project. Initially, he used Claude Code with auto-approve turned off, reviewing every file edit and terminal command before execution. "The results were good. The code was clean. But I was doing most of the thinking and half the typing. The agent was a fancy autocomplete with better suggestions," Johnson explains.

A turning point came when he read about "on-the-loop" versus "in-the-loop" human-AI collaboration: "I was micromanaging because I didn't trust the agent to do the right thing. And I didn't trust the agent because there was nothing forcing it to do the right thing."

With proper testing, static analysis, and architectural patterns in place, Johnson could shift his role from writer to curator:

• Defining patterns
• Reviewing test specs
• Reviewing output
• Updating the harness
• Making strategic decisions

This represents a significant shift in how developers might collaborate with AI tools, moving beyond simple code completion to more substantial architectural partnership.

Open Source Security Concerns

Back in the UK, the National Health Service's decision to close nearly all their Open Source repositories sparked notable groans. The stated reason was security concerns related to LLMs, but this approach raises questions about effectiveness and transparency.

Government Data Services (GDS), the highly-regarded IT enablers in the UK government, have published their position on this matter: "Moving code from public to private as a substitute for investment in secure-by-design delivery, ownership and remediation is a warning sign because it reduces sharing and scrutiny, can slow coordinated improvement across government and suppliers, and does not remove the underlying weaknesses in a running service."

Terence Eden memorably sums up the bureaucratic attitude: "Within the UK's Civil Service you occasionally hear the expression 'being invited to a meeting without biscuits'. It implies a rather frosty discussion without any of the polite niceties of a normal meeting."

Cognitive Challenges with AI Agents

Adam Tornhill has joined a growing group of developers working with LLMs who are experiencing cognitive endurance problems. "One of the big wins with agents is that they let us stay with the higher-level problem for longer. We get less sidetracked by details, dependency cleanup, and similar secondary tasks that used to break concentration. But there is a cost we are still underestimating," Tornhill explains.

He describes agentic coding as "mentally expensive," noting that he can usually sustain the pace for a couple of hours before needing a break. "The pace is simply too intense. And based on conversations with other engineers, I do not think I am alone in that."

Working with AI agents like The Genie increases decision density, which places significant strain on cognitive resources. Tornhill responds by:

• Keeping agent tasks small
• Automating everything possible
• Accepting that he won't know every line of code as long as good verification mechanisms are in place

Notably, he avoids the current hype of running multiple agents in parallel: "I cannot even think about twenty meaningful things to build, and even less so about the resulting cognitive tax of the likely interruptions. It's exactly the wrong thing to even consider. At least for humans."

Tornhill also emphasizes the importance of activities outside intense programming sessions, particularly learning about the domain that the software supports, suggesting a balanced approach to AI-assisted development.

Generational Perspectives on AI

Karl Bode observes that younger generations are increasingly unhappy with the tech oligarchy and their products. "The thing is the kids aren't stupid. They see the field clearly. They see the difference between what's being sold to them by tech companies, the press, and commencement speakers, and what they have repeatedly seen with their own eyes," Bode writes.

He notes that Gen Z has watched tech leaders navigate "scandal after scandal, hype cycle after hype cycle, steadily enshittifying everything they touch along the way." This has led to significant skepticism:

• Around 50% of Gen Z believes AI's benefits don't counterbalance the risks (up 11 percentage points in just the last year)
• Eight out of ten believe that using AI makes the process of actual learning more difficult

Bode connects this skepticism to broader concerns about entering a worsening world, suggesting that "This rage that could have marked political and social consequences."

Economic Impact of AI

The Economist newspaper presents a more nuanced perspective on AI's economic impact. Historically, major technological advances haven't led to significant unemployment or drops in wages, with the closest example being the original industrial revolution in 19th Century Britain. During that period, there was wage stagnation but also a massive population increase, from 4.5 million to 12 million.

The newspaper suggests we'll only understand AI's full consequences when a recession hits, as this is when most unproductive jobs tend to be flushed out of the system.

A separate analysis indicates AI is already affecting graduate hiring. The least exposed quintile of subjects saw employment rates fall by 1.5% over the last couple of years, while the most exposed quintile's drop was 6.6%.

Regulatory Challenges

Lawfare expresses disappointment with the US Government's efforts to regulate AI. The White House recently planned to have leaders from major AI companies sign an executive order on AI and cybersecurity, but canceled the ceremony roughly three hours before it was scheduled to take place.

The proposed regulations were characterized as mild, including valuable measures to harden defenses against cyber threats. However, the cancellation suggests that "voluntary, in other words, isn't the floor of frontier AI policy in this administration; it's the ceiling."

One significant problem is the distinct lack of governmental expertise in AI and software. "Too much is being decided at the whims of the tech oligarchy, there isn't any attempt to engage in the broader issues at hand," the analysis notes.

This creates a dangerous situation where the impact of AI is so substantial that being too far behind in regulation poses real risks. As Lawfare points out, killing the order just leaves in place what has been described as an "opaque and essentially lawless" alternative: government access happening through back channels, on terms set case by case, with no stable rules at all.

Pithy Quotes

From social media:

• Lorin Hochstein: "Metaphor debt" is when all of your metaphors involve the concept of "debt" because you can't think of any other metaphors anymore.
• Daniel Terhorst-North: If a vegan crossfit fan is using Claude to write Rust, which thing do they tell you first?

A Call for Technical Expertise in Government

In an unusual political endorsement, Martin Fowler expresses support for Beth Anders-Beck, who is running for Congress in Massachusetts's 6th district. "Beth has a long background in software development (including developing the notion of Forest and Desert), so would introduce expertise that Congress desperately needs," Fowler writes.

Fowler, who has known Anders-Beck for decades, praises her "intelligence, judgment, and ability to work with others," noting that "Congress doesn't deserve Beth, but it does need her." This endorsement highlights the growing recognition that technical expertise is essential for effective governance in an increasingly digital world.

As AI continues to transform software development and society at large, these diverse perspectives offer valuable insights into the challenges and opportunities ahead. The tension between technological advancement and human needs, between innovation and regulation, and between efficiency and cognitive well-being will continue to shape our technological future.